Lifeboat Foundation

Professor Jeongho Kwak’s from the Department of Electrical Engineering and Computer Science at DGIST has developed a learning model and resource optimization technology that combines accuracy and efficiency for 6G vision services. This technology is expected to be utilized to address the high levels of computing power and complex learning models required by 6G vision services.

6G mobile vision services are associated with innovative technologies such as augmented reality (AR) and autonomous driving, which are receiving significant attention in modern society. These services enable quick capturing of videos and images, and efficient understanding of their content through deep learning-based models.

However, this requires high-performance processors (GPUs) and accurate learning models. Previous technologies treated learning models and computing/networking resources as separate entities, failing to optimize performance and mobile device resource utilization.

The streaming audio giant’s suite of recommendation tools has grown over the years: Spotify Home feed, Discover Weekly, Blend, Daylist, and Made for You Mixes. And in recent years, there have been signs that it is working. According to data released by Spotify at its 2022 Investor Day, artist discoveries every month on Spotify had reached 22 billion, up from 10 billion in 2018, “and we’re nowhere near done,” the company stated at that time.

Over the past decade or more, Spotify has been investing in AI and, in particular, in machine learning. Its recently launched AI DJ may be its biggest bet yet that technology will allow subscribers to better personalize listening sessions and discover new music. The AI DJ mimics the vibe of radio by announcing the names of songs and lead-in to tracks, something aimed in part to help ease listeners into extending out of their comfort zones. An existing pain point for AI algorithms — which can be excellent at giving listeners what it knows they already like — is anticipating when you want to break out of that comfort zone.

Artificial intelligence excels at sorting through information and detecting patterns or trends. But these machine learning algorithms need to be trained with large amounts of data first.

As researchers explore potential applications for AI, they have found scenarios where AI could be really useful—such as analyzing X-ray image data to look for evidence of rare conditions or detecting a rare fish species caught on a commercial fishing boat—but there’s not enough data to accurately train the algorithms.

Jenq-Neng Hwang, University of Washington professor of electrical and computer and engineering, specializes in these issues. For example, Hwang and his team developed a method that teaches AI to monitor how many distinct poses a baby can achieve throughout the day. There are limited training datasets of babies, which meant the researchers had to create a unique pipeline to make their algorithm accurate and useful.

In this episode, recorded during the 2024 Abundance360 Summit, Ray, Geoffrey, and Peter debate whether AI will become sentient, what consciousness constitutes, and if AI should have rights.

Ray Kurzweil, an American inventor and futurist, is a pioneer in artificial intelligence. He has contributed significantly to OCR, text-to-speech, and speech recognition technologies. He is the author of numerous books on AI and the future of technology and has received the National Medal of Technology and Innovation, among other honors. At Google, Kurzweil focuses on machine learning and language processing, driving advancements in technology and human potential.

Continue reading “Ray Kurzweil & Geoff Hinton Debate the Future of AI | EP #95” »

Strategy could stop an overdose or produce an antidote to a poison.

Scientists have created a method to produce 3D full-color holographic images using smartphone screens instead of lasers. This innovative technique, with additional advancements, holds the potential for augmented or virtual reality displays.

Whether augmented and virtual reality displays are being used for gaming, education, or other applications, incorporating 3D displays can create a more realistic and interactive user experience.

“Although holography techniques can create a very real-looking 3D representation of objects, traditional approaches aren’t practical because they rely on laser sources,” said research team leader Ryoichi Horisaki, from The University of Tokyo in Japan. “Lasers emit coherent light that is easy to control, but they make the system complex, expensive, and potentially harmful to the eyes.”

Carnegie Mellon University (CMU) researchers have developed H₂O – Human2HumanOid – a reinforcement learning-based framework that allows a full-sized humanoid robot to be teleoperated by a human in real-time using only an RGB camera. Which begs the question: will manual labor soon be performed remotely?

A teleoperated humanoid robot allows for the performance of complex tasks that are – at least at this stage – too complex for a robot to perform independently. But achieving whole-body control of human-sized humanoids to replicate our movements in real-time is a challenging task. That’s where reinforcement learning (RL) comes in.

Continue reading “How blue-collar workers will train the humanoids that take their jobs” »

Language models often need more exposure to fruitful mistakes during training, hindering their ability to anticipate consequences beyond the next token. LMs must improve their capacity for complex decision-making, planning, and reasoning. Transformer-based models struggle with planning due to error snowballing and difficulty in lookahead tasks. While some efforts have integrated symbolic search algorithms to address these issues, they merely supplement language models during inference. Yet, enabling language models to search for training could facilitate self-improvement, fostering more adaptable strategies to tackle challenges like error compounding and look-ahead tasks.

Researchers from Stanford University, MIT, and Harvey Mudd have devised a method to teach language models how to search and backtrack by representing the search process as a serialized string, Stream of Search (SoS). They proposed a unified language for search, demonstrated through the game of Countdown. Pretraining a transformer-based language model on streams of search increased accuracy by 25%, while further finetuning with policy improvement methods led to solving 36% of previously unsolved problems. This showcases that language models can learn to solve problems via search, self-improve, and discover new strategies autonomously.

Recent studies integrate language models into search and planning systems, employing them to generate and assess potential actions or states. These methods utilize symbolic search algorithms like BFS or DFS for exploration strategy. However, LMs primarily serve for inference, needing improved reasoning ability. Conversely, in-context demonstrations illustrate search procedures using language, enabling the LM to conduct tree searches accordingly. Yet, these methods are limited by the demonstrated procedures. Process supervision involves training an external verifier model to provide detailed feedback for LM training, outperforming outcome supervision but requiring extensive labeled data.

A newly developed AI method can calculate a fundamental problem in quantum chemistry: Schrödinger’s Equation. The technique could calculate the ground state of the Schrödinger equation in quantum chemistry.

Predicting molecules’ chemical and physical properties by relying on their atoms’ arrangement in space is the main goal of quantum chemistry. This can be achieved by solving the Schrödinger equation, but in practice, this is extremely difficult.